Files
homelab-frontend/internal/sessions/sandbox.go
T
Admin Bot 16ddd6d4ad
CI / CI (pull_request) Successful in 2m8s
feat(sessions): sandbox resource — per-session k8s Pod+PVC+Service (§4.15.4)
New X-Resource: sandbox on X-Service: sessions

POST ?session_id=  → CreateOrResume
  1. Ensure PVC workspace-session-<id> (Longhorn 1Gi)
  2. If pod already Running → return DNS (resume path, idempotent)
  3. Ensure pi-session-sandbox NetworkPolicy (shared, created once)
  4. Create Pod pi-session-<id> + Service pi-session-<id>:8787
  5. Wait up to 30s for Pod Ready
  6. Fire-and-forget: merge pod_name/pvc_name/svc_dns/last_active_at
     into session memory_entity summary

DELETE ?session_id= → delete Pod + Service + PVC

Idle reaper goroutine (5min interval, configurable via
SANDBOX_REAPER_INTERVAL_MINUTES): lists pi-session-* pods,
deletes Succeeded/Failed ones, preserves PVCs for resume.

Security: runAsNonRoot/1000, readOnlyRootFilesystem, drop ALL
caps, RuntimeDefault seccomp, automountServiceAccountToken:false.
Resources: 500m/512Mi req, 1/1Gi limit.
NetworkPolicy: egress → DNS + api:8080 only; ingress ← poimen-frontend.

RBAC: Role session-sandbox-manager + RoleBinding in poimen ns.
7 unit tests — all pass (mock k8s server).

Image: forgejo.riotpiao.com/riotpiao-poimen/poimen-pi-session-runtime:latest
2026-09-28 19:53:37 +08:00

493 lines
16 KiB
Go

// Package sessions — SandboxManager provisions per-session k8s resources:
//
// PVC workspace-session-<uuid> Longhorn 1Gi
// Pod pi-session-<uuid> single container, /workspace <- PVC
// Svc pi-session-<uuid> ClusterIP :8787
// NetPol pi-session-sandbox shared for all pi-session pods (created once)
//
// All k8s API calls use raw HTTP + in-cluster SA token (no client-go).
// Same pattern as internal/webhook/promoter.go.
package sessions
import (
"bytes"
"context"
"crypto/tls"
"crypto/x509"
"encoding/json"
"fmt"
"io"
"log"
"net/http"
"os"
"strings"
"time"
)
const (
sandboxNamespace = "poimen"
sandboxImage = "forgejo.riotpiao.com/riotpiao-poimen/poimen-pi-session-runtime:latest"
sandboxPort = 8787
defaultPVCSize = "1Gi"
defaultStorageClass = "longhorn"
defaultReadyTimeout = 30 * time.Second
defaultReaperInterval = 5 * time.Minute
sandboxSATokenPath = "/var/run/secrets/kubernetes.io/serviceaccount/token"
sandboxCACertPath = "/var/run/secrets/kubernetes.io/serviceaccount/ca.crt"
)
// SandboxManager manages the k8s lifecycle of pi-session Pods, PVCs, Services.
type SandboxManager struct {
k8sHost string
k8sClient *http.Client
pvcSize string
readyTimeout time.Duration
}
// SandboxResult is returned by CreateOrResume.
type SandboxResult struct {
SessionID string `json:"session_id"`
PodName string `json:"pod_name"`
PVCName string `json:"pvc_name"`
SvcDNS string `json:"svc_dns"`
Resumed bool `json:"resumed"` // true = pod already existed
}
// NewSandboxManager creates a SandboxManager from env + in-cluster SA files.
func NewSandboxManager() *SandboxManager {
host := os.Getenv("K8S_API_HOST")
if host == "" {
host = "https://kubernetes.default.svc"
}
pvcSize := os.Getenv("SANDBOX_PVC_SIZE")
if pvcSize == "" {
pvcSize = defaultPVCSize
}
caCert, _ := os.ReadFile(sandboxCACertPath)
return &SandboxManager{
k8sHost: host,
k8sClient: buildSandboxK8sClient(caCert),
pvcSize: pvcSize,
readyTimeout: defaultReadyTimeout,
}
}
func buildSandboxK8sClient(caCert []byte) *http.Client {
transport := &http.Transport{}
if len(caCert) > 0 {
pool := x509.NewCertPool()
pool.AppendCertsFromPEM(caCert)
transport.TLSClientConfig = &tls.Config{RootCAs: pool}
} else {
log.Printf("sandbox: WARNING no SA CA cert at %s — using InsecureSkipVerify", sandboxCACertPath)
transport.TLSClientConfig = &tls.Config{InsecureSkipVerify: true} //nolint:gosec
}
return &http.Client{Timeout: 30 * time.Second, Transport: transport}
}
func sandboxSAToken() string {
b, err := os.ReadFile(sandboxSATokenPath)
if err != nil {
log.Printf("sandbox: WARNING failed to read SA token %s: %v", sandboxSATokenPath, err)
return ""
}
return strings.TrimSpace(string(b))
}
func pvcName(sessionID string) string { return "workspace-session-" + sessionID }
func podName(sessionID string) string { return "pi-session-" + sessionID }
func svcName(sessionID string) string { return "pi-session-" + sessionID }
func svcDNS(sessionID string) string {
return fmt.Sprintf("%s.%s.svc.cluster.local:%d", svcName(sessionID), sandboxNamespace, sandboxPort)
}
// ── public API ────────────────────────────────────────────────────────────
// CreateOrResume provisions or resumes a sandbox for sessionID.
// Idempotent: if the pod is already Running, returns its DNS immediately.
func (s *SandboxManager) CreateOrResume(ctx context.Context, sessionID string) (*SandboxResult, error) {
res := &SandboxResult{
SessionID: sessionID,
PodName: podName(sessionID),
PVCName: pvcName(sessionID),
SvcDNS: svcDNS(sessionID),
}
// 1. Ensure PVC exists
if err := s.ensurePVC(ctx, sessionID); err != nil {
return nil, fmt.Errorf("ensure PVC: %w", err)
}
// 2. Check if pod already running (resume case)
phase, err := s.getPodPhase(ctx, sessionID)
if err != nil {
return nil, fmt.Errorf("check pod: %w", err)
}
if phase == "Running" {
res.Resumed = true
return res, nil
}
// 3. Ensure NetworkPolicy for all pi-session pods exists
if err := s.ensureNetworkPolicy(ctx); err != nil {
log.Printf("sandbox: WARNING failed to ensure network policy: %v", err)
}
// 4. Create Pod + Service
if err := s.createPod(ctx, sessionID); err != nil {
return nil, fmt.Errorf("create pod: %w", err)
}
if err := s.ensureService(ctx, sessionID); err != nil {
return nil, fmt.Errorf("ensure service: %w", err)
}
// 5. Wait for Ready
if err := s.waitPodReady(ctx, sessionID); err != nil {
return nil, fmt.Errorf("wait ready: %w", err)
}
return res, nil
}
// Delete removes the Pod, Service, and PVC for sessionID.
func (s *SandboxManager) Delete(ctx context.Context, sessionID string) error {
var errs []string
for _, fn := range []func(context.Context, string) error{
s.deletePod, s.deleteService, s.deletePVC,
} {
if err := fn(ctx, sessionID); err != nil && !isNotFound(err) {
errs = append(errs, err.Error())
}
}
if len(errs) > 0 {
return fmt.Errorf("delete sandbox: %s", strings.Join(errs, "; "))
}
return nil
}
// ReapIdlePods lists all pi-session-* pods and deletes those that have exited
// (Succeeded/Failed) — bridge.js self-exits after idle timeout.
// The PVC is preserved so the next CreateOrResume reattaches the same workspace.
func (s *SandboxManager) ReapIdlePods(ctx context.Context) {
pods, err := s.listSandboxPods(ctx)
if err != nil {
log.Printf("sandbox reaper: list pods: %v", err)
return
}
for _, podAny := range pods {
pod, _ := podAny.(map[string]any)
phase, _ := pod["status"].(map[string]any)["phase"].(string)
name, _ := pod["metadata"].(map[string]any)["name"].(string)
if phase == "Succeeded" || phase == "Failed" {
log.Printf("sandbox reaper: deleting exited pod %s (phase=%s)", name, phase)
if err := s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+name); err != nil && !isNotFound(err) {
log.Printf("sandbox reaper: delete pod %s: %v", name, err)
}
}
}
}
// ── k8s resource helpers ─────────────────────────────────────────────────
func (s *SandboxManager) ensurePVC(ctx context.Context, sessionID string) error {
name := pvcName(sessionID)
_, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/persistentvolumeclaims/"+name)
if err == nil {
return nil // already exists
}
if !isNotFound(err) {
return err
}
// Create
body := map[string]any{
"apiVersion": "v1",
"kind": "PersistentVolumeClaim",
"metadata": map[string]any{
"name": name, "namespace": sandboxNamespace,
"labels": map[string]string{"app": "pi-session", "session-id": sessionID},
},
"spec": map[string]any{
"accessModes": []string{"ReadWriteOnce"},
"storageClassName": defaultStorageClass,
"resources": map[string]any{
"requests": map[string]string{"storage": s.pvcSize},
},
},
}
return s.k8sCreate(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/persistentvolumeclaims", body)
}
func (s *SandboxManager) getPodPhase(ctx context.Context, sessionID string) (string, error) {
pod, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+podName(sessionID))
if isNotFound(err) {
return "", nil
}
if err != nil {
return "", err
}
phase, _ := pod["status"].(map[string]any)["phase"].(string)
return phase, nil
}
func (s *SandboxManager) createPod(ctx context.Context, sessionID string) error {
body := map[string]any{
"apiVersion": "v1",
"kind": "Pod",
"metadata": map[string]any{
"name": podName(sessionID), "namespace": sandboxNamespace,
"labels": map[string]string{"app": "pi-session", "session-id": sessionID},
},
"spec": map[string]any{
"restartPolicy": "Never",
"automountServiceAccountToken": false,
"securityContext": map[string]any{
"runAsNonRoot": true,
"runAsUser": int64(1000),
"seccompProfile": map[string]string{"type": "RuntimeDefault"},
},
"containers": []map[string]any{{
"name": "pi-session",
"image": sandboxImage,
"ports": []map[string]any{{"containerPort": sandboxPort}},
"securityContext": map[string]any{
"readOnlyRootFilesystem": true,
"allowPrivilegeEscalation": false,
"capabilities": map[string]any{"drop": []string{"ALL"}},
},
"resources": map[string]any{
"requests": map[string]string{"cpu": "500m", "memory": "512Mi"},
"limits": map[string]string{"cpu": "1", "memory": "1Gi"},
},
"readinessProbe": map[string]any{
"tcpSocket": map[string]any{"port": sandboxPort},
"initialDelaySeconds": 3,
"periodSeconds": 2,
"failureThreshold": 15,
},
"volumeMounts": []map[string]any{
{"name": "workspace", "mountPath": "/workspace"},
{"name": "tmp", "mountPath": "/tmp"},
},
}},
"volumes": []map[string]any{
{"name": "workspace", "persistentVolumeClaim": map[string]string{"claimName": pvcName(sessionID)}},
{"name": "tmp", "emptyDir": map[string]any{}},
},
},
}
return s.k8sCreate(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods", body)
}
func (s *SandboxManager) ensureService(ctx context.Context, sessionID string) error {
name := svcName(sessionID)
_, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/services/"+name)
if err == nil {
return nil
}
if !isNotFound(err) {
return err
}
body := map[string]any{
"apiVersion": "v1",
"kind": "Service",
"metadata": map[string]any{
"name": name, "namespace": sandboxNamespace,
"labels": map[string]string{"app": "pi-session", "session-id": sessionID},
},
"spec": map[string]any{
"type": "ClusterIP",
"selector": map[string]string{"app": "pi-session", "session-id": sessionID},
"ports": []map[string]any{{
"port": sandboxPort, "targetPort": sandboxPort, "protocol": "TCP",
}},
},
}
return s.k8sCreate(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/services", body)
}
func (s *SandboxManager) ensureNetworkPolicy(ctx context.Context) error {
const npName = "pi-session-sandbox"
_, err := s.k8sGet(ctx, "/apis/networking.k8s.io/v1/namespaces/"+sandboxNamespace+"/networkpolicies/"+npName)
if err == nil {
return nil // already exists
}
if !isNotFound(err) {
return err
}
body := map[string]any{
"apiVersion": "networking.k8s.io/v1",
"kind": "NetworkPolicy",
"metadata": map[string]any{
"name": npName, "namespace": sandboxNamespace,
"labels": map[string]string{"app": "pi-session"},
},
"spec": map[string]any{
"podSelector": map[string]any{"matchLabels": map[string]string{"app": "pi-session"}},
"policyTypes": []string{"Ingress", "Egress"},
"ingress": []map[string]any{{
"from": []map[string]any{{
"podSelector": map[string]any{"matchLabels": map[string]string{"app": "poimen-frontend"}},
}},
"ports": []map[string]any{{"protocol": "TCP", "port": sandboxPort}},
}},
"egress": []map[string]any{
// DNS
{
"to": []map[string]any{{"namespaceSelector": map[string]any{"matchLabels": map[string]string{"kubernetes.io/metadata.name": "kube-system"}}}},
"ports": []map[string]any{{"protocol": "UDP", "port": 53}, {"protocol": "TCP", "port": 53}},
},
// api-gateway (LLM inference)
{
"to": []map[string]any{{"namespaceSelector": map[string]any{"matchLabels": map[string]string{"kubernetes.io/metadata.name": "api"}}}},
"ports": []map[string]any{{"protocol": "TCP", "port": 8080}},
},
},
},
}
return s.k8sCreate(ctx, "/apis/networking.k8s.io/v1/namespaces/"+sandboxNamespace+"/networkpolicies", body)
}
func (s *SandboxManager) waitPodReady(ctx context.Context, sessionID string) error {
deadline := time.Now().Add(s.readyTimeout)
for time.Now().Before(deadline) {
pod, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+podName(sessionID))
if err != nil {
return fmt.Errorf("poll pod: %w", err)
}
status, _ := pod["status"].(map[string]any)
phase, _ := status["phase"].(string)
if phase == "Failed" || phase == "Succeeded" {
return fmt.Errorf("pod entered terminal phase %s before ready", phase)
}
if phase == "Running" {
for _, cond := range toSlice(status["conditions"]) {
if c, ok := cond.(map[string]any); ok {
if c["type"] == "Ready" && c["status"] == "True" {
return nil
}
}
}
}
select {
case <-time.After(2 * time.Second):
case <-ctx.Done():
return ctx.Err()
}
}
return fmt.Errorf("pod not ready after %v", s.readyTimeout)
}
func (s *SandboxManager) deletePod(ctx context.Context, sessionID string) error {
return s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods/"+podName(sessionID))
}
func (s *SandboxManager) deleteService(ctx context.Context, sessionID string) error {
return s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/services/"+svcName(sessionID))
}
func (s *SandboxManager) deletePVC(ctx context.Context, sessionID string) error {
return s.k8sDelete(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/persistentvolumeclaims/"+pvcName(sessionID))
}
func (s *SandboxManager) listSandboxPods(ctx context.Context) ([]any, error) {
body, err := s.k8sGet(ctx, "/api/v1/namespaces/"+sandboxNamespace+"/pods?labelSelector=app%3Dpi-session")
if err != nil {
return nil, err
}
items, _ := body["items"].([]any)
return items, nil
}
// ── raw k8s HTTP ─────────────────────────────────────────────────────────
func (s *SandboxManager) k8sGet(ctx context.Context, path string) (map[string]any, error) {
req, err := http.NewRequestWithContext(ctx, http.MethodGet, s.k8sHost+path, nil)
if err != nil {
return nil, err
}
req.Header.Set("Authorization", "Bearer "+sandboxSAToken())
resp, err := s.k8sClient.Do(req)
if err != nil {
return nil, fmt.Errorf("k8s GET %s: %w", path, err)
}
defer resp.Body.Close()
if resp.StatusCode == 404 {
return nil, &k8sNotFound{path: path}
}
if resp.StatusCode >= 400 {
b, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
return nil, fmt.Errorf("k8s GET %s returned %d: %s", path, resp.StatusCode, b)
}
var result map[string]any
if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
return nil, fmt.Errorf("decode: %w", err)
}
return result, nil
}
func (s *SandboxManager) k8sCreate(ctx context.Context, path string, body any) error {
payload, err := json.Marshal(body)
if err != nil {
return fmt.Errorf("marshal: %w", err)
}
req, err := http.NewRequestWithContext(ctx, http.MethodPost, s.k8sHost+path, bytes.NewReader(payload))
if err != nil {
return err
}
req.Header.Set("Authorization", "Bearer "+sandboxSAToken())
req.Header.Set("Content-Type", "application/json")
resp, err := s.k8sClient.Do(req)
if err != nil {
return fmt.Errorf("k8s POST %s: %w", path, err)
}
defer resp.Body.Close()
if resp.StatusCode == 409 { // AlreadyExists — idempotent
return nil
}
if resp.StatusCode < 200 || resp.StatusCode >= 300 {
b, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
return fmt.Errorf("k8s POST %s returned %d: %s", path, resp.StatusCode, b)
}
return nil
}
func (s *SandboxManager) k8sDelete(ctx context.Context, path string) error {
req, err := http.NewRequestWithContext(ctx, http.MethodDelete, s.k8sHost+path, nil)
if err != nil {
return err
}
req.Header.Set("Authorization", "Bearer "+sandboxSAToken())
resp, err := s.k8sClient.Do(req)
if err != nil {
return fmt.Errorf("k8s DELETE %s: %w", path, err)
}
defer resp.Body.Close()
if resp.StatusCode == 404 {
return &k8sNotFound{path: path}
}
if resp.StatusCode >= 400 {
b, _ := io.ReadAll(io.LimitReader(resp.Body, 512))
return fmt.Errorf("k8s DELETE %s returned %d: %s", path, resp.StatusCode, b)
}
return nil
}
// ── error helpers ─────────────────────────────────────────────────────────
type k8sNotFound struct{ path string }
func (e *k8sNotFound) Error() string { return "not found: " + e.path }
func isNotFound(err error) bool {
if err == nil {
return false
}
_, ok := err.(*k8sNotFound)
return ok
}
func toSlice(v any) []any {
if s, ok := v.([]any); ok {
return s
}
return nil
}